| Benchmark | Gemini 3.1 Pro | o3 |
|---|---|---|
| AA Agentic Index | 23 | 36.1 |
| AA Intelligence | 47.7 | 31.1 |
| AA-LCR | 79 | 73.3 |
| AA-Omniscience | 32.9 | -15.3 |
| arc_agi_1 | 98 | 60.8 |
| ARC-AGI-2 | 77.1 | 6.5 |
| arena_vision | 1280 | 1217 |
| Artificial Analysis Coding Index | 68.8 | 38.4 |
| browsecomp | 85.9 | 49.7 |
| critpt | 17.7 | 1.1 |
| Fortress | 29.8 | 16 |
| frontiermath_tier_4 | 16.7 | 2.1 |
| gdpval | 23.3 | 12.8 |
| GPQA Diamond | 94.3 | 83.3 |
| HLE | 51.4 | 20.3 |
| IFBench | 77.1 | 71.4 |
| LiveCodeBench | 91.7 | 84.7 |
| MMLU-Pro | 91 | 85 |
| MMMU-Pro | 83 | 76.4 |
| multichallenge | 71.4 | 60.4 |
| MultiNRC | 64.7 | 45.5 |
| OmniScience Accuracy | 55.3 | 38.6 |
| OmniScience Non-Hallucination | 50.1 | 12.9 |
| scicode | 59 | 41 |
| simplebench | 79.6 | 53.1 |
| simpleqa_verified | 77.3 | 53 |
| strongreject | 98 | 97 |
| SWE-bench Verified | 80.6 | 69.1 |
| Terminal-Bench Hard | 68.5 | 37.1 |
| VisualToolBench | 29 | 13.7 |
| τ²-Bench Telecom (AA run) | 99.3 | 80.7 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.